Re: [PATCH] tracing/filters: allow on-the-fly filter switching

[Tom Zanussi <tzanussi@gmail.com>]

On Tue, 2009-04-14 at 22:56 +0200, Frederic Weisbecker wrote:
> On Mon, Apr 13, 2009 at 03:17:50AM -0500, Tom Zanussi wrote:

[...]

> 
> 
> BTW, this issue might be already present before this patch.
> What happens if:
> 
> 
> 	T1				T2
> 
> event_filter_read() {
>   filter_print_preds() {
>     for (i = 0; i < n_preds; i++) {
> 	pred = preds[i];
> 				    event_filter_write() {
> 					filter_disable_preds();
> 					filter_clear_preds() {
> 					  kfree(pred->field_name);
> 	field_name = pred->field_name;
> 	// CRASH!!!
> 
> 
> You need a mutex to protect these two callbacks.
> It would also protect concurrent calls to event_filter_write(),
> which would result in random.
> 

Yeah, Li Zefan had already fixed this, but it wasn't included in this
patch.  Looks like he'll be resubmitting that part...

> 
> 

[...]

> Also, is filter_add_pred() supposed to be available for in-kernel
> uses by other tracers or something?

No, the current callers were the only ones I'd planned on (it's not
static because code in trace_events.c needs to call it).  But, do you
see a use for it by other tracers?

> If this is planned, the locking could be even deeper than my comments.
> 
> Other than these comments:
> 
> Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
> 
> Thanks!
> 

Thanks for reviewing it!

Tom